CSPSAT Projects and My Research Topicsresearch.nii.ac.jp/il/web/10years-symposium/slides/soh.pdf ·...

Outline

.

......

CSPSAT Projects andMy Research Topics

Takehide Soh

March 18, 2014 at NII

Takehide Soh CSPSAT Projects and My Research Topics

Outline

.. Self Introduction

Takehide SOH

...1 2008.04-2011.11SOKENDAI, Ph.D., in Inoue Lab.Studied applications of SAT technologies(solving 2SPP and pathway analyses)

...2 2011.11-2012.03TRIC, Post-doc., in Inoue Lab.Continuously studied pathway analyses

...3 2012.04-CurrentKobe Univ., Assist. Prof., in Tamura Lab.Studying SAT-based CP System writtenin Scala (Scarab) and its applications.


Outline

.. Table of Contents

Overview of CSPSAT and CSPSAT2 Project

My Research Topics on CSPSAT

My Research Topics on CSPSAT2


Outline

.. CSPSAT and CSPSAT2 Projects

.Overview..

......

CSPSAT project started on 2008 and is followed by runningCSPSAT2 project started on 2012.

The goal of projects is the research and development ofenhanced SAT technologies as high-performance inferenceengines for hard problems.

.Developed Software..

......

SAT-based CSP Solvers Sugar, Azucar

Max-SAT and PB Solvers QMaxSat, PBSugar

SAT Solvers GlueMiniSat, SCSat

Domain Specific Languages for CP Scarab, Copris and more...

.

...... Introduce members and some researches in CSPSAT projects.


Outline

.. Members of CSPSAT

Yamanashi

− Koji Iwanuma

− Hidetomo Nabeshima

− Yoshitaka Yamamoto

NII

− Katsumi Inoue

Waseda

− Kazunori Ueda

Kobe

− Naoyuki Tamura

− Katsutoshi Hirayama

− Mutsunori Banbara

Kyushu

− Ryuzo Hasegawa

− Makoto Yokoo

− Hiroshi Fujita

− Miyuki Koshimura


Outline

..Sugar (Tamura et al.)SAT-based CSP Solver

Based on a new encoding named Order Encoding [Tamura etal., Constraints 2009]

Winner of 2008 and 2009 CSP Solver Competitions inGLOBAL categories

Closed three instances (j7-per0-0, j8-per0-1, j8-per10-2) ofOpen-Shop Scheduling Problem [Tamura et al., CP2006]

There is also another version Azucar using Compact OrderEncoding [Tanjo et al., SAT 2012].


Outline

..GlueMiniSat (Nabeshima et al.)A Fast SAT Solver with An Aggressive Acquiring Strategy of Glue Clause

Won several prizes of SAT competitions of 2011 and 2013 inApplication, UNSAT category.Strong for UNSAT


Outline

..Dynamic CSP/SAT (Hatano, Hirayama et al.)Solving a Sequence of CSP/SAT Instances with Decision Change Costs

Find a sequence of solutions that minimizes the sum ofdecision change costs for a given sequence of CSP/SATinstances.

Lagrangian Decomposition works very well in terms ofsolution quality reached within a fixed amount of time[Hatano and Hirayama, IJCAI 2011].


Outline

..

ASP as a Modeling Language for CB-CTT (Banbaraet al.)

Proposed ASP-based method provides 175 best known bounds.

Methods Authors # of BestsASP-based M. Banbara 175Tabu Search (110)

A. Schaerf 111Hybrid Methods (1)

S. Abdullah &Other

H. Turabieh30

Tabu Search Z. Lu & J. Hao 24SAT-based Barcelogic Team 19Mathematical Programming A. Phillips 18Mathematical Programming G. Lach 5Local Search T. Muller 5Simulated Annealing SaTT group 3Very Large Neighborhood Search A. Kiefer 2Simulated Annealing M. Muehlenthaler 1Hybrid Methods Khalid & Salwan 1


Outline

..My Research Topics on CSPSATTwo-Dimensional Strip Packing (Soh and Inoue et al.)

Optimum Packing of HT08

Find a minimum height to pack all given rectangles in a fixedwidth strip.

Closed 29 instances [Soh et al., FI, 2010]


Outline

..My Research Topics on CSPSATPrediction of Gene Knockout Effects by Minimal Model Generation

Glycolysis Pentose phosphate

gapA pgk enofbaA

fbaA

gapA

pgk

eno dld

tpiA

serA

hyi

agpAlpha-GCL-6-PGLC

3-p-OH-pyr

TARTRONATE-S-ALD

glxR garR

ATP

NADP+

nadk

Source metabolites (MS): β-D-glucose-6-phosphate, NAD+

Target metabolite (MT): pyruvate

Four essential genes confirmed by KEIO collection

Genes predicted by our method

G6P

PYR

G6P

PYR

pgm

pgi

zwf

gpmA gpmB ytjC

pfkA pfkB

ATP

ADP

pykF pykA

ADP

ADP ATP

gnd

pgl

Lactate

edd

6PG

NADP+

ATP ADP

NAD+

NAD+

Gluconolactone-6P

NAD+

fbp

fbaB

talA talB

E4P

S7P

ppsA pflB

ldhA

tktAtktB

tktAtktB

rpe

rpiA rpiB

eda

Essential Genes of E. coli Predicted by a SAT-based Analysis Method[Soh and Inoue et al., Intl. J. Advances in Life Sciences, 2012]


Outline

.. My Research Topics on CSPSAT2: Motivation

.Researches I studied until 2012..

......

Solving Two-Dimensional Strip Packing Problem

Proposing a new solving method for an existing problem

Analyses for Metabolic Pathways

Focusing on a Biological problem (Gene knockout effects)Represent it as a new problem on bipartite digraphsProposing a solving method

Both researches use SAT technologies and need trial and erroron propositional modeling.

.

...... There are general demands of an useful workbench on SAT.

This is the motivation for the development of Scarab.


Outline

..My Research Topics on CSPSAT2: ScarabTight Integrated System with SAT Solvers

Scarab is a prototyping tool for developing SAT-basedConstraint Programming (CP) systems.

It consists of 1) CP Domain-Specific Language, 2) API of CSPsolver, 3) SAT encoding module, and 4) API of SAT solvers.It uses Order Encoding and Sat4j in default.

It is developed to be an expressive, efficient, customizable,and portable workbench.


Outline


Scarab is a prototyping tool for developing SAT-basedConstraint Programming (CP) systems.It consists of 1) CP Domain-Specific Language, 2) API of CSPsolver, 3) SAT encoding module, and 4) API of SAT solvers.It uses Order Encoding and Sat4j in default.

Sat4j

Scarab

DSLSAT Solver

API

Encoder

Decoder

CSP Solver

API

Implemented by 800 lines of Scala



Outline



ScarabProgram

Sat4j

ScarabDSL

SAT Solver API

Encoder

Decoder

CSP SolverAPI

(DSL+Scala)



Outline



ScarabProgram

Sat4j

CSPobject

ScarabDSL

SAT Solver API

Encoder

Decoder

CSP SolverAPI

(DSL+Scala)



Outline



It consists of 1) CP Domain-Specific Language, 2) API of CSPsolver, 3) SAT encoding module, and 4) API of SAT solvers.

It uses Order Encoding and Sat4j in default.

ScarabProgram

Sat4j

CSPobject

ScarabDSL

SAT Solver API

Encoder

Decoder

MAP

SATobject

CSP SolverAPI

(DSL+Scala)



Outline



It consists of 1) CP Domain-Specific Language, 2) API of CSPsolver, 3) SAT encoding module, and 4) API of SAT solvers.

It uses Order Encoding and Sat4j in default.

ScarabProgram

Sat4j

CSPobject

ScarabDSL

SAT Solver API

Encoder

CSPsolution

SATsolutionDecoder

MAP

SATobject

CSP SolverAPI

(DSL+Scala)



Outline

..My Research Topics on CSPSAT2: ScarabApplications using Scarab

.

......

We are now developing several applications using Scarab and theyshow good performances compared to other systems.

CP

U T

ime

(se

c.)

#Solved Instances

For Post-Enrollment Course Timetabling

0

100

200

300

400

500

0 50 100 150 200 250 300 350 400 450

Tim

e (

se

c)

#Solved

[Velev ’09]Lingeling-ats + Seq. Counter

Lingeling-ats + TotalizerLKH 2.0.7

Sat4j + TotalizerMinisat2.2 + TotalizerGlucose3 + Totalizer

Minisat2.2 + Seq. CounterGlucose3 + Seq. Counter

Sat4j + Seq. CounterSat4j + Native BC

For Hamiltonian Cycle Problem


Outline

.. Web Page for Scarab

http://kix.istc.kobe-u.ac.jp/~soh/scarab/


Outline

.. Web Page for CSPSAT2

http://www.edu.kobe-u.ac.jp/istc-tamlab/cspsat/en/


CSPSAT Projects and My Research Topicsresearch.nii.ac.jp/il/web/10years-symposium/slides/soh.pdf ·...

Documents

Transcript of CSPSAT Projects and My Research Topicsresearch.nii.ac.jp/il/web/10years-symposium/slides/soh.pdf ·...